Preliminaries for Model Predictive Control course

James B. Rawlings
Department of Chemical and Biological Engineering
University of WisconsinMadison

Insitut f
ur Systemtheorie und Regelungstechnik
Universitat Stuttgart
Stuttgart, Germany
June 20, 2011

Stuttgart June 2011

MPC short course

1 / 46

Outline

Introduction

Stability, equilibria, invariant sets

Lyapunov function theory

Disturbances and robust stability

Basic MPC problem and nominal stability

Stuttgart June 2011

MPC short course

2 / 46

Predictive control
Reconcile the past
11
00
00
11
0
1
0
1
0
1

11
00
00
11
00
11

MH Estimate
Measurement

0
1
0
1
1
0
1
0
1
0
0
1
11
0
1
0
0
1
0
1
0
1 1
0
01
1
0
0
1
0
0
1
0
01
0 0
01
1
0
1
1 1
0
1
1
0
1
0

sensors
y

Forecast the future

1
0
1
0

1
0
0
1
0
1
0
1

1
0
1
0
1
0
0
1

1
0
0
1
1
0
1
0

1
0
0
1

1
0
1
0
1
0
1
0
01
1
0
01
0
11
0
11
0
0
0
1

1
0
0
1

1
0
0
1
10
0
1
0
1
0
1

0
11
0
0
1
0
1
1
0
1
0

1
0
0
1

1
0
0
1
1
0
1
0

1
0
0
1

1
0
1
0

1
0
1
0

1
0
0
1

Forecast
MPC control

1
0
0
0
0
0
0
0
0
0
0
0
0
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
0
1

1
0
0
1
1
0

actuators
u
time

|ysp g (x, u)|2Q + |usp u|2R dt

min
u(t)

x = f (x, u)
x(0) = x0

(given)

y = g (x, u)
Stuttgart June 2011

MPC short course

3 / 46

State estimation
Reconcile the past
11
00
00
11
0
1
0
1
0
1

sensors
y

0
1
0
1
1
0
1
0
1
0
0
1
11
0
1
0
0
1
0
1
0
1 1
0
01
1
0
0
1
0
0
1
0
01
0 0
01
1
0
1
1 1
0
1
1
0
1
0

1
0
0
1

Forecast the future

11
00
00
11
00
11

MH Estimate
Measurement

1
0
1
0

1
0
0
1
0
1
0
1

1
0
1
0
1
0
0
1

1
0
0
1
1
0
1
0

1
0
0
1

1
0
1
0
1
0
1
0
0
1
0
01
1
0
1
1
0
11
0
0
0
1

1
0
0
1
10
0
1
0
1
0
1
1
0
0
1
1
0
1
0

0
11
0
0
1
0
1
1
0
1
0

1
0
0
1

1
0
0
1

1
0
1
0

1
0
1
0

1
0
0
1

Forecast
MPC control

1
0
0
0
0
0
0
0
0
0
0
0
0
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
01
1
0
1

1
0
0
1
1
0

actuators
u
t

min
x0 ,w (t) T

time

|y g (x, u)|2R + |x f (x, u)|2Q dt

x = f (x, u) + w

(process noise)

y = g (x, u) + v

(measurement noise)

Stuttgart June 2011

MPC short course

4 / 46

System model1
We consider systems of the form
x + = f (x, u)
where the state x lies in Rn and the control (input) u lies in Rm ;
In this formulation x and u denote, respectively, the current state and
control, and x + the successor state.
We assume in the sequel that the function f : Rn Rm Rn is
continuous.
Let
(k; x, u)
denote the solution of x + = f (x, u) at time k if the initial state is
x(0) = x and the control sequence is u = {u(0), u(1), u(2), . . .};
The solution exists and is unique.
1

Most of this preliminary material is taken from Rawlings and Mayne (2009,
Appendix B). Downloadable from www.che.wisc.edu/~jbraw/mpc.
Stuttgart June 2011

MPC short course

5 / 46

Existence of solutions to model

If a state-feedback control law u = (x) has been chosen, the
closed-loop system is described by x + = f (x, (x)).
Let (k; x, ()) denote the solution of this difference equation at
time k if the initial state at time 0 is x(0) = x; the solution exists and
is unique (even if () is discontinuous).
If () is not continuous, as may be the case when () is a model
predictive control (MPC) law, then f ((), ()) may not be continuous.
In this case we assume that f ((), ()) is locally bounded.

Definition 1 (Locally bounded)

A function f : X X is locally bounded if, for any x X , there exists a
neighborhood N of x such that f (N ) is a bounded set, i.e., if there exists
a M > 0 such that |f (x)| M for all x N .

Stuttgart June 2011

MPC short course

6 / 46

Stability and equilibrium point

We would like to be sure that the controlled system is stable, i.e., that
small perturbations of the initial state do not cause large variations in the
subsequent behavior of the system, and that the state converges to a
desired state or, if this is impossible due to disturbances, to a desired set
of states.
If convergence to a specified state, x say, is sought, it is desirable for this
state to be an equilibrium point:

Definition 2 (Equilibrium point)

A point x is an equilibrium point of x + = f (x) if x(0) = x implies
x(k) = (k; x ) = x for all k 0. Hence x is an equilibrium point if it
satisfies
x = f (x )

Stuttgart June 2011

MPC short course

7 / 46

Positive invariant set

In other situations, for example when studying the stability properties of

an oscillator, convergence to a specified closed set A Rn is sought.
If convergence to a set A is sought, it is desirable for the set A to be
positive invariant:

Definition 3 (Positive invariant set)

A set A is positive invariant for the system x + = f (x) if x A implies
f (x) A.
Clearly, any solution of x + = f (x) with initial state in A, remains in A.
The (closed) set A = {x } consisting of a (single) equilibrium point is a
special case; x A (x = x ) implies f (x) A (f (x) = x ).

Stuttgart June 2011

MPC short course

8 / 46

Distance to a set; set addition

Define distance from point x to set A

|x|A := inf |x z|
zA

If A = {x }, then |x|A = |x x | which reduces to |x| when x = 0.

Set addition: A B := {a + b | a A, b B}.

Stuttgart June 2011

MPC short course

9 / 46

K, K , KL, and PD functions

Definition 4
A function : R0 R0 belongs to class K if it is continuous, zero
at zero, and strictly increasing;
: R0 R0 belongs to class K if it is a class K and unbounded
((s) as s ).
A function : R0 I0 R0 belongs to class KL if it is
continuous and if, for each t 0, (, t) is a class K function and for
each s 0, (s, ) is nonincreasing and satisfies limt (s, t) = 0.
A function : R R0 belongs to class PD (is positive definite) if it
is continuous and positive everywhere except at the origin.

Stuttgart June 2011

MPC short course

10 / 46

Some useful properties of K functions

The following useful properties of these functions are established in Khalil

(2002, Lemma 4.2):
if 1 () and 2 () are K functions (K functions), then 11 () and
(1 2 )() := 1 (2 ()) are K functions (K functions).
Moreover, if 1 () and 2 () are K functions and () is a KL
function, then (r , s) = 1 ((2 (r ), s)) is a KL function.

Stuttgart June 2011

MPC short course

11 / 46

Stability Definitions

Definition 5 (Various forms of stability (constrained))

Suppose X Rn is positive invariant for x + = f (x), that A is closed and
positive invariant for x + = f (x), and that A lies in the interior of X . Then
A is
1

locally stable in X if, for each > 0, there exists a = () > 0 such
that x X (A B), implies |(i; x)|A < for all i I0 .a

locally attractive in X if there exists a > 0 such that

x X (A B) implies |(i; x)|A 0 as i .

attractive in X if |(i; x)|A 0 as i for all x X .

B denotes the unit ball in Rn .

Stuttgart June 2011

MPC short course

12 / 46

Stability Definitions (cont.)

Definition 5 (Various forms of stability (constrained))

4

locally asymptotically stable in X if it is locally stable in X and locally

attractive in X .

asymptotically stable with a region of attraction X if it is locally

stable in X and attractive in X .

locally exponentially stable with a region of attraction X if there exist

> 0, c > 0, and (0, 1) such that x X (A B) implies
|(i; x)|A c|x|A i for all i I0 .

exponentially stable with a region of attraction X if there exists a

c > 0 and a (0, 1) such that |(i; x)|A c|x|A i for all x X ,
all i I0 .

Stuttgart June 2011

MPC short course

13 / 46

Asymptotic stability stronger definition

An alternative, stronger definition of asymptotic stability is becoming
popular. Ill call it the KL version.

Definition 6 (Asymptotic stability (constrained KL version))

Suppose X Rn is positive invariant for x + = f (x), that A is closed and
positive invariant for x + = f (x), and that A lies in the interior of X . The
set A is asymptotically stable with a region of attraction X for x + = f (x)
if there exists a KL function () such that, for each x X
|(i; x)|A (|x|A , i)

i I0

(1)

See Teel and Zaccarian (2006) and the Notes on Recent MPC
Literature link on: www.che.wisc.edu/~jbraw/mpc for further
discussion of the differences in the two definitions.
If f () is continuous, the two definitions are equivalent.
Stuttgart June 2011

MPC short course

14 / 46

Lyapunov function
Definition 7 (Lyapunov function)
A function V : Rn R0 is said to be a Lyapunov function for the system
x + = f (x) and set A if there exist functions i K , i = 1, 2 and
3 PD such that for any x Rn ,
V (x) 1 (|x|A )

(2)

V (x) 2 (|x|A )

(3)

V (f (x)) V (x) 3 (|x|A )

(4)

If V () satisfies (2)(4) for all x X where X A is a positive invariant

set for x + = f (x), then V () is said to be a Lyapunov function in X for
the system x + = f (x) and set A.

Stuttgart June 2011

MPC short course

15 / 46

Some fine print

Remark 8
It is shown in Jiang and Wang (2002), Lemma 2.8, that, under the
assumption that f () is continuous, we can always assume that 3 () in (4)
is a K function.

Stuttgart June 2011

MPC short course

16 / 46

Lyapunov function Asymptotic stability

The existence of a Lyapunov function is a sufficient condition for global

asymptotic stability as shown in the next result which we prove under the
assumption, common in MPC, that 3 () is K function.

Theorem 9 (Lyapunov function and GAS)

Suppose V () is a Lyapunov function for x + = f (x) and set A with 3 ()
a K function. Then A is globally asymptotically stable.

Stuttgart June 2011

MPC short course

17 / 46

Converse Lyapunov theorem Asymptotic stability

Theorem 9 merely provides a sufficient condition for global asymptotic
stability that might be thought to be conservative. The next result, a
converse stability theorem by Jiang and Wang (2002), establishes necessity
under a stronger hypothesis, namely that f () is continuous rather than
locally bounded.

Theorem 10 (Converse theorem for asymptotic stability)

Let A be compact and f () continuous. Suppose that the set A is globally
asymptotically stable for the system x + = f (x). Then there exists a
smooth Lyapunov function for the system x + = f (x) and set A.
A proof of this result is given in Jiang and Wang (2002), Theorem 1,
part 3.

Stuttgart June 2011

MPC short course

18 / 46

Asymptotic stability Constrained case

Theorem 11 (Lyapunov function for asymptotic stability (constrained

case))
Suppose X Rn is positive invariant for x + = f (x), that A is closed and
positive invariant for x + = f (x), and that A lies in the interior of X . If
there exists a Lyapunov function in X for the system x + = f (x) and set A
with 3 () a K function, then A is asymptotically stable for x + = f (x)
with a region of attraction X .

Stuttgart June 2011

MPC short course

19 / 46

Exponential stability Constrained case

Theorem 12 (Lyapunov function for exponential stability)

Suppose X Rn is positive invariant for x + = f (x), that A is closed and
positive invariant for x + = f (x), and that A lies in the interior of X . If
there exists V : Rn R0 satisfying the following properties for all x X
a1 |x|A V (x) a2 |x|A
V (f (x)) V (x) a3 |x|A
in which a1 , a2 , a3 , > 0, then A is exponentially stable for x + = f (x)
with a region of attraction X .

Stuttgart June 2011

MPC short course

20 / 46

Converse theorem for exponential stability

Exercise B.3: A converse theorem for exponential stability
a

Assume that the origin is globally exponentially stable (GES) for the
system
x + = f (x)
in which f is Lipschitz continuous. Show that there exists a Lipschitz
continuous Lyapunov function V () for the system satisfying for all
x Rn
a1 |x| V (x) a2 |x|
V (f (x)) V (x) a3 |x|

in which a1 , a2 , a3 , > 0.
Hint: Consider summing the solution |(i; x)| on i as a candidate
Lyapunov function V (x).
Establish also that in the Lyapunov function defined above, any > 0
is valid, and the constant a3 can be chosen as large as one wishes.
Stuttgart June 2011

MPC short course

21 / 46

Robust Stability
We turn now to stability conditions for systems subject to bounded
disturbances (not vanishingly small) and described by
x + = f (x, w )

(5)

where the disturbance w lies in the compact set W.

This system may equivalently be described by the difference inclusion
x + F (x)

(6)

where the set

F (x) := {f (x, w ) | w W}
Let S(x) denote the set of all solutions of (5) or (6) with initial state x.

Stuttgart June 2011

MPC short course

22 / 46

Positive invariant sets

We require, in the sequel, that the set A is positive invariant for (5) (or
for x + F (x)):

Definition 13 (Positive invariance with disturbances)

The set A is positive invariant for x + = f (x, w ), w W if x A implies
f (x, w ) A for all w W; it is positive invariant for x + F (x) if x A
implies F (x) A.
Clearly the two definitions are equivalent; A is positive invariant for
x + = f (x, w ), w W, if and only if it is positive invariant for x + F (x).
In Definitions 14-16, we use positive invariant to denote positive
invariant for x + = f (x, w ), w W or for x + F (x).

Stuttgart June 2011

MPC short course

23 / 46

Stability and attraction

Definition 14 (Local stability (disturbances))

The closed positive invariant set A is locally stable for x + = f (x, w ),
w W (or for x + F (x)) if, for all > 0, there exists a > 0 such that,
for each x satisfying |x|A < , each solution S(x) satisfies |(i)|A <
for all i I0 .

Definition 15 (Global attraction (disturbances))

The closed positive invariant set A is globally attractive for the system
x + = f (x, w ), w W (or for x + F (x)) if, for each x Rn , each
solution () S(x) satisfies |(i)|A 0 as i .

Stuttgart June 2011

MPC short course

24 / 46

Asymptotic stability

Definition 16 (GAS (disturbances))

The closed positive invariant set A is globally asymptotically stable for
x + = f (x, w ), w W (or for x + F (x)) if it is locally stable and globally
attractive.
An alternative definition of global asymptotic stability of A for
x + = f (x, w ), w W, if A is compact, is the existence of a KL function
() such that for each x Rn , each S(x) satisfies
|(i)|A (|x|A , i) for all i I0 .

Stuttgart June 2011

MPC short course

25 / 46

Lyapunov function
To cope with disturbances we require a modified definition of a Lyapunov
function.

Definition 17 (Lyapunov function (disturbances))

A function V : Rn R0 is said to be a Lyapunov function for the system
x + = f (x, w ), w W (or for x + F (x)) and set A if there exist
functions i K , i = 1, 2 and 3 PD such that for any x Rn ,
V (x) 1 (|x|A )

(7)

V (x) 2 (|x|A )

(8)

sup V (z) V (x) 3 (|x|A )

(9)

zF (x)

Stuttgart June 2011

MPC short course

26 / 46

GAS (disturbances)

Inequality 9 ensures V (f (x, w )) V (x) 3 (|x|A ) for all w W. The

existence of a Lyapunov function for the system x + F (x) and set A is a
sufficient condition for A to be globally asymptotically stable for
x + F (x) as shown in the next result.

Theorem 18 (Lyapunov function for GAS (disturbances))

Suppose V () is a Lyapunov function for x + = f (x, w ), w W (or for
x + F (x)) and set A with 3 () a K function. Then A is globally
asymptotically stable for x + = f (x, w ), w W (or for x + F (x)).

Stuttgart June 2011

MPC short course

27 / 46

Input-to-State Stability
In input-to-state stability (Sontag and Wang, 1995; Jiang and Wang,
2001) we seek a bound on the state in terms of a uniform bound on the
disturbance sequence w := {w (0), w (1), . . .}. Let kk denote the usual `
norm for sequences, i.e., kwk := supk0 |w (k)|.

Definition 19 (Input-to-state stable (ISS))

The system x + = f (x, w ) is (globally) input-to-state stable (ISS) if there
exists a KL function () and a K function () such that, for each
x Rn , and each disturbance sequence w = {w (0), w (1), . . .} in `
|(i; x, wi )| (|x| , i) + (kwi k)
for all i I0 , where (i; x, wi ) is the solution, at time i, if the initial state
is x at time 0 and the input sequence is wi := {w (0), w (1), . . . , w (i 1)}.

Stuttgart June 2011

MPC short course

28 / 46

The basic nonlinear, constrained MPC problem

The system model is

x + = f (x, u)

(10)

Both state and input are subject to constraints

x(k) X ,

u(k) U

for all k I0

Given an integer N (referred to as the finite horizon), and an input

sequence u of length N, u = {u(0), u(1), . . . , u(N 1)}, let (k; x, u)
denote the solution of (10) at time k for a given initial state x(0) = x.
Terminal constraint (and penalty)
(N; x, u) Xf X

Stuttgart June 2011

MPC short course

29 / 46

Feasible sets
The set of feasible initial states and associated control sequences
ZN = {(x, u) | u(k) U, (k; x, u) X for all k I0:N1 ,
and (N; x, u) Xf }
and Xf X is the feasible terminal set.
The set of feasible initial states is
XN = {x Rn | u UN such that (x, u) ZN }

(11)

For each x XN , the corresponding set of feasible input sequences is

UN (x) = {u | (x, u) ZN }

Stuttgart June 2011

MPC short course

30 / 46

Cost function and control problem

For any state x Rn and input sequence u UN , we define

VN (x, u) =

N1
X

`((k; x, u), u(k)) + Vf ((N; x, u))

k=0

`(x, u) is the stage cost; Vf (x(N)) is the terminal cost

Consider the finite horizon optimal control problem
PN (x) :

Stuttgart June 2011

min VN (x, u)

uUN

MPC short course

31 / 46

Control law and closed-loop system

The control law is

N (x) = u 0 (0; x)
The optimum may not be unique; then N () is a point-to-set map
Closed-loop system
x + = f (x, N (x))

difference equation

x + f (x, N (x))

difference inclusion

Nominal closed-loop stability question; is the origin stable?

If yes, what is the region of attraction? All of XN ?

Stuttgart June 2011

MPC short course

32 / 46

Inherent robustness of the nominal controller

Consider a process disturbance d, x + = f (x, (x)) + d

A measurement disturbance xm = x + e
Nominal controller with disturbance
x + f (x, N (xm )) + d
x + f (x, N (x + e)) + d
x + F (x, w )

w = (d, e)

Robust stability; is the system x + F (x, w ) input-to-state stable

considering w = (d, e) as the input.

Stuttgart June 2011

MPC short course

33 / 46

Basic MPC assumptions

Assumption 20 (Continuity of system and cost)

The functions f : Rn Rm Rn , ` : Rn Rm R0 and
Vf : Rn R0 are continuous, f (0, 0) = 0, `(0, 0) = 0, and Vf (0) = 0.

Assumption 21 (Properties of constraint sets)

The set U is compact and contains the origin. The sets X and Xf are
closed and contain the origin in their interiors, Xf X.

Stuttgart June 2011

MPC short course

34 / 46

Basic MPC assumptions

Assumption 22 (Basic stability assumption)
For any x Xf there exists u := f (x) U such that f (x, u) Xf and
Vf (f (x, u)) + `(x, u) Vf (x).
Note: understanding this requirement created a big research challenge for
the development of nonlinear MPC. Credit the celebrated quasi-infinite
horizon work of Chen and Allgower (1998) for cracking this problem.

Assumption 23 (Bounds on stage and terminal costs)

The stage cost `() and the terminal cost Vf () satisfy
`(x, u) 1 (|x|)
Vf (x) 2 (|x|)

x XN , u U
x Xf

in which 1 () and 2 () are K functions

Stuttgart June 2011

MPC short course

35 / 46

Optimal MPC cost function as Lyapunov function

We show that the optimal cost VN0 () is a Lyapunov function for the
closed-loop system. We require three properties.
Lower bound.
VN0 (x) 1 (|x|)

for all

x XN

Given the definition of VN (x, u) as a sum of stage costs, we have using

Assumption 23
VN (x, u) `(x, u(0; x)) 1 (|x|)

for all

x XN , u UN

so the first property is established.

Stuttgart June 2011

MPC short course

36 / 46

MPC cost function as Lyapunov function cost decrease

Next we require the cost decrease
VN0 (f (x, N (x)) VN0 (x) 3 (|x|)

for all

x XN

At state x XN , consider the optimal sequence

u0 (x) = {u(0; x), u(1; x), . . . , u(N 1; x)}, and generate a candidate
sequence for the successor state, x + := f (x, N (x))
= {u(1; x), u(2; x), . . . , u(N 1; x), f (x(N))}
u
with x(N) := (N; x, u). This candidate is feasible for x + because Xf is
control invariant under control law f () (Assumption 22).
The cost is
) = VN0 (x) `(x, u(0; x))
VN (x + , u
Vf (x(N)) + `(x(N), f (x(N))) + Vf (f (x(N), f (x(N))))
Stuttgart June 2011

MPC short course

37 / 46

Cost decrease (cont.)

But by Assumption 22
Vf (f (x, f (x))) + `(x, f (x)) Vf (x) for all

x Xf

so we have that
) VN0 (x) `(x, u(0; x))
VN (x + , u
The optimal cost is certainly no worse, giving
VN0 (x + ) VN0 (x) `(x, u(0; x))
VN0 (x + ) VN0 (x) 1 (|x|)

for all

x XN

which is the desired cost decrease with the choice 3 () = 1 ().

Stuttgart June 2011

MPC short course

38 / 46

Upper bound
Finally we require the upper bound.
VN0 (x) 2 (|x|)

for all

x XN

Surprisingly, this one turns out to be the most involved.

First, we have the bound from Assumption 23
Vf (x) 2 (|x|)

for all

x Xf

Next we show that VN0 (x) Vf (x) for x Xf , N 1.

Consider N = 1,
V10 (x) = min{`(x, u) + Vf (f (x, u)) | f (x, u) Xf }
uU

Vf (x)

x Xf

Stuttgart June 2011

MPC short course

39 / 46

Dynamic programming recursion

Next consider N = 2, and optimal control law 2 ()
V20 (x) = min{`(x, u) + V10 (f (x, u)) | f (x, u) X1 }
uU

= `(x, 2 (x)) + V10 (f (x, 2 (x)))

x X2

x X2

`(x, 1 (x)) + V10 (f (x, 1 (x))) x X1

`(x, 1 (x)) + Vf (f (x, 1 (x)))
= V10 (x)

x X1

x X1

Therefore
V20 (x) Vf (x) x Xf
Continuing this recursion gives for all N 1
VN0 (x) Vf (x)

Stuttgart June 2011

x Xf

MPC short course

40 / 46

Extending the upper bound from Xf to XN

Question: When can we extend this bound from Xf to the (possibly
unbounded!) set XN ? Recall that VN0 () is not necessarily continuous.
Answer: A function can be upper bounded by a K function if and
only if it is locally bounded.2
We know from continuity of f () (Assumption 20) that VN (x, u) is a
continuous function, hence locally bounded, and therefore so is
VN0 (x).
Therefore, there exists () K such that
VN0 (x) (|x|)

for all

x XN

Be aware that the MPC literature has been confused about the
requirements for this last result.
2

See Proposition 10 of Notes on Recent MPC Literature link on:

www.che.wisc.edu/~jbraw/mpc. Thanks also to Andy Teel.
Stuttgart June 2011

MPC short course

41 / 46

Asymptotic stability of constrained nonlinear MPC

Why you want a Lyapunov function
We have established that the optimal cost VN0 () is a Lyapunov
function on XN for the closed-loop system.
Therefore, the origin is asymptotically stable (KL version) with region
of attraction XN .
We can also establish robust stability, but lets do that later.
If we strengthen the properties of `(), we can strengthen the
conclusion to exponential stability.
Notice the essential role that VN0 () plays in the stability analysis of
MPC.
In economic MPC we lose this Lyapunov function and have to work to
bring it back.

Stuttgart June 2011

MPC short course

42 / 46

Recommended exercises

Stability definitions. Example 2.3

Lyapunov functions. Exercise B.2B.4.4
Dynamic programming. Exercise C.1C.2.4
MPC stability results. Theorem 7 and Example 1.3
Exercises 2.11, 2.14, 2.154

Notes on Recent MPC Literature link on: www.che.wisc.edu/~jbraw/mpc.

Rawlings and Mayne (2009, Appendices B and C). Downloadable from
www.che.wisc.edu/~jbraw/mpc.
4

Stuttgart June 2011

MPC short course

43 / 46

Further Reading I
H. Chen and F. Allg
ower. A quasi-infinite horizon nonlinear model
predictive control scheme with guaranteed stability. Automatica, 34(10):
12051217, 1998.
Z.-P. Jiang and Y. Wang. Input-to-state stability for discrete-time
nonlinear systems. Automatica, 37:857869, 2001.
Z.-P. Jiang and Y. Wang. A converse Lyapunov theorem for discrete-time
systems with disturbances. Sys. Cont. Let., 45:4958, 2002.
H. K. Khalil. Nonlinear Systems. Prentice-Hall, Upper Saddle River, NJ,
third edition, 2002.
J. B. Rawlings and D. Q. Mayne. Model Predictive Control: Theory and
Design. Nob Hill Publishing, Madison, WI, 2009. 576 pages, ISBN
978-0-9759377-0-9.
E. D. Sontag and Y. Wang. On the characterization of the input to state
stability property. Sys. Cont. Let., 24:351359, 1995.
Stuttgart June 2011

MPC short course

44 / 46

Further Reading II

A. R. Teel and L. Zaccarian. On uniformity in definitions of global

asymptotic stability for time-varying nonlinear systems. Automatica, 42:
22192222, 2006.

Stuttgart June 2011

MPC short course

45 / 46

Acknowledgments

The author is indebted to Luo Ji of the University of Wisconsin and

Ganzhou Wang of the Universitat Stuttgart for their help in organizing the
material and preparing the overheads.

Stuttgart June 2011

MPC short course

46 / 46